This study used a convergent mixed methods design, in which quantitative and qualitative data were collected and analyzed concurrently and integrated during interpretation [31]. This feasibility study is part of a broader research program focused on developing speech-based cognitive assessment methods using audio-recorded patient-clinician interactions and follow-up calls in HHC settings. These approaches leverage speech-derived features to support the early detection of cognitive impairment [17]. This study was conducted at a large nonprofit HHC agency in New York City, which serves about 50,000 patients per year, over 30% of whom are Black. Data collection was conducted across 4 boroughs, including Manhattan, Brooklyn, Queens, and the Bronx, from January 10, 2024, to September 1, 2024. Both HHC patients and clinicians were included in this study.

Eligible patients included are as follows: (1) self-identified Black adults aged 60 years and older who spoke English, were able to communicate independently with clinicians, and had the capacity to provide informed consent; (2) individuals with MCI or ED, defined by an International Classification of Diseases, 10th Revision (ICD-10) diagnosis (G31.84) or a Clinical Dementia Rating (CDR) score of 0 to 1 and a Montreal Cognitive Assessment (MoCA) score of 18 to 25, with neurologist confirmation; (3) cognitively normal individuals with no documented cognitive impairment, a CDR score of 0, and a MoCA score of ≥26; and (4) participants with sufficient hearing and vision to complete cognitive assessments. Patients were excluded if they had severe cognitive impairment (CDR 2‐3; MoCA 0‐9), major psychiatric or neurological conditions, recent substance abuse, severe communication impairment, unstable or serious medical conditions such as severe pain, active cancer within 2 years, significant sensory impairments, or other conditions limiting participation.

Eligible clinicians were licensed registered nurses, physical therapists, or medical social workers employed by the participating HHC agency, with at least 1 year of HHC experience. All participating clinicians had established clinical relationships with the enrolled patients.

The study was approved by the institutional review board of the participating agency (approval: E23-01). Information on informed consent, privacy, confidentiality, and compensation is further detailed in the following sections.

To assess patient and clinician experiences with the study procedures, we collected qualitative data through structured follow-up interviews and open-ended questionnaires.

Figure 1 outlines the recruitment and participation flow across the 4 components of the study. In component 1, we attempted contact with 246 patients by phone. Of these, we were unable to reach 40, and 135 declined participation due to reasons such as health issues (47/135, 35%), lack of interest (40/135, 30%), caregiver refusal (12/135, 9%), or skepticism about research, artificial intelligence, or audio recording (10/135, 7%). The reasons listed represent the most common reasons identified and are not intended to be an exhaustive list. Ultimately, 71 patients verbally consented to participate. Before cognitive testing in component 2, 11 withdrew, leaving 60 who provided written consent and completed the assessments. Of these, 5 were deemed ineligible due to disqualifying health conditions (n=3) or severe cognitive impairment (n=2) based on the screening results. All 55 eligible patients were invited to participate in component 3 (follow-up phone calls); 48 (87%) agreed, while 7 (13%) declined due to unreturned calls (n=4), caregiver coordination issues (n=2), or a disconnected phone line (n=1). In component 4, 54 of 55 (98%) eligible patients consented to audio record their HHC visits, with 1 declining due to discomfort sharing personal details. Overall, the recruitment process yielded a consent rate of 24% (60/246) from initial outreach and strong participation rates across cognitive assessment (60/71, 85%), follow-up calls (48/55, 87%), and audio-recorded visits (54/55, 98%), demonstrating the feasibility of engaging Black patients in audio-based research within HHC settings. Ten participating clinicians provided feedback, including 5 registered nurses, 4 physical therapists, and 1 medical social worker. Participants included 2 Black, 5 White, 2 Hispanic, and 1 Asian clinician. Most (7/10, 70%) reported 1 to 5 years of experience, while 3 (30%) reported more than 5 years.

Table 1 compares the characteristics of those who provided written consent (60/71, 84.5%) to those who did not (146/206, 71%), a group that includes both initial decliners and patients who withdrew prior to written consent. No significant differences were observed in gender, age, insurance type, living situation, social isolation, anxiety levels, or activity of daily living scores. However, nonconsenting patients were more likely to require assistance reading medication labels (P=.04), experience frequent pain interference (P=.01), take 5 or more medications (P=.01), and have a diagnosis of heart disease (P=.05). They also reported marginally higher levels of exhaustion (P=.08). These findings suggest that greater medical burden and functional limitations may hinder research participation among HHC Black patients.

Integration of quantitative and qualitative findings showed that lower consent rates among patients with greater medical burdens were consistent with reported barriers such as pain and competing health demands. In contrast, high participation across study components aligned with patients’ motivation to contribute to research and positive study experiences. Both patients and clinicians reported that audio recording was acceptable and minimally disruptive, supporting the feasibility of integrating speech-based data collection into routine HHC (Table 2).

This study examines the recruitment and participation process for an audio-recorded speech study in an HHC setting. Of the patients initially contacted, approximately 1 in 4 agreed to participate, demonstrating a moderate recruitment rate among an older adult population, which is particularly notable given our exclusive focus on older Black patients receiving HHC, a population often underrepresented in cognitive impairment research. Importantly, we found that patients with low literacy and greater medical complexity (eg, polypharmacy, frequent pain interference, and heart disease) were more likely to decline participation, highlighting the need for research strategies to account for or identify better ways to include individuals with a greater burden of chronic conditions. Despite these challenges, both clinicians and patients who participated reported positive attitudes toward audio recording, with minimal disruption to standard care. This underscores the potential for innovative, noninvasive diagnostic methods, such as speech-based assessments, to be integrated into routine HHC visits for earlier detection of MCI-ED.

Recruiting older adults, particularly those with MCI-ED, presents unique challenges and often leads to lower participation rates in research [39,40]. Numerous studies have reported the underrepresentation of older adults in clinical trials, especially among those in the oldest age groups [41,42]. In contrast, our study achieved relatively balanced enrollment across age groups, suggesting that targeted outreach and tailored messaging can enhance the inclusion of older adults at varying levels of cognitive and medical complexity. Nevertheless, although our overall consent rate was slightly lower than the 38% reported by other studies that focused on dementia research [39], it remains noteworthy in the context of reaching Black patients with multiple comorbidities [43,44].

Older adults with MCI-ED often face multiple barriers to research participation, such as lack of interest, caregiver discouragement, or feeling overwhelmed by their conditions. A systematic review on the recruitment and retention of older adults in clinical research found that lack of interest and external discouragement were among the most common reasons for refusal [45]. Consistent with these findings, our study also identified these 2 barriers. Additionally, our findings demonstrated that patients with greater medical complexity (eg, polypharmacy and frequent pain interference) were more likely to decline, consistent with research showing that poor health status can impede enrollment [46,47]. Several factors may explain these differences. Patients with higher symptom burden and functional limitations may experience fatigue, pain, or competing care demands, reducing their capacity or willingness to engage in research activities. In addition, managing multiple medications and chronic conditions may increase cognitive and logistical burdens, making participation in additional study procedures less appealing.

Importantly, integrating qualitative findings provides additional insight into these patterns. While patients with greater medical complexity were less likely to participate, those who enrolled were often motivated by altruism and a desire to understand their cognitive health. Qualitative data also suggest that some barriers, such as pain or time constraints, may be modifiable. For example, several patients who initially declined audio recording during home visits subsequently expressed interest in participating following follow-up contact, suggesting that recruitment decisions may be dynamic rather than fixed. In addition, participants consistently reported that audio recording of patient-clinician encounters was unobtrusive and did not interfere with care. Together, these findings suggest that although greater disease burden may limit initial participation, targeted strategies, such as leaving contact information and inviting patients to reach out if they later express interest, along with clear communication and reassurance about study procedures, may improve recruitment among higher-risk individuals.

While quantitative findings indicated that patients with greater medical burden (eg, polypharmacy and pain interference) were more likely to decline participation, qualitative data provide important context for these patterns. Many participants reported that their decision to enroll was driven by altruistic motivations (eg, contributing to dementia research) and personal interest in understanding their cognitive health. Notably, qualitative findings also suggest that initial barriers such as pain or time constraints may be modifiable; for example, some patients who initially declined later agreed to participate following follow-up contact. In addition, participants who enrolled generally reported positive experiences and minimal burden, particularly regarding audio recording, which was perceived as unobtrusive. Together, these findings suggest that while higher medical burden may limit initial participation, targeted strategies such as follow-up engagement, clear communication, and reassurance about study procedures may help overcome initial reluctance and improve recruitment. A key outcome of this study was the generally positive reception of audio recording by both patients and nurses, which supported high participation rates in the latter components (eg, audio-recorded home visits). Previous work suggests that embedding research procedures into existing care routines, rather than adding separate appointments, enhances participant engagement and retention [48,49]. In our study, face-to-face interactions during home visits likely fostered trust and comfort, aligning with research showing that relational continuity in HHC can improve study adherence [49,50]. Audio recording was perceived as unobtrusive and straightforward to integrate into routine workflows, affirming earlier findings on the feasibility of speech-based data collection in HHC settings [30].

Future research should prioritize the development of tailored recruitment strategies that address the diverse needs and health complexities of older adults with MCI-ED, particularly those with severe medical conditions who are often underrepresented in research. Customized approaches that take into account both medical and personal circumstances such as personalized communication, flexible participation options, and targeted outreach can overcome specific barriers. Additionally, clear and accessible communication materials that explain the study’s purpose, benefits, and safety measures can help build trust, reduce skepticism, and minimize caregiver refusals. Such approaches could reduce both logistical and psychological barriers, ultimately improving the diversity and representativeness of study samples.

The application of audio recordings for speech analysis has shown significant promise as a tool for the early detection of MCI-ED [16,51]. The positive feedback from study participants highlights the feasibility and potential for integrating this method into routine HHC diagnostics. Future research should further investigate and validate the effectiveness of audio-recorded speech data in supporting the early identification of MCI-ED. By enabling early detection and timely interventions, this approach could potentially slow or even halt the progression of cognitive decline in this vulnerable population, thereby improving patient outcomes and quality of life.

Despite offering valuable insights, our study faces several limitations that warrant consideration. First, a substantial number of patients declined participation during recruitment, which may have introduced selection bias and limited the generalizability of the findings, particularly among individuals with more intensive health needs who may be underrepresented. As a result, the sample may not fully capture the experiences of the most severely affected segments of the MCI-ED population, who are often the primary targets for early diagnostic tools and interventions. However, given that this step was intended to assess the feasibility of the approach, it reflects real-world engagement with an understudied population. Second, we did not collect detailed demographic information for clinicians and only captured limited characteristics (eg, professional role and years of experience), which limits the interpretation of clinician perspectives. Third, because the study was conducted within a technology-supported environment, findings may not be transferable to settings with limited access to technology or among populations more skeptical of artificial intelligence-based research. Fourth, although the study by design specifically focused on including older Black patients, barriers faced by other racial and ethnic groups in this setting could not be determined. Nonetheless, the approach taken in this study will inform subsequent steps, including engaging more groups of older adults in New York and elsewhere. Finally, while most participants were comfortable with audio recording, some expressed privacy concerns that may have led to hesitation in sharing sensitive information and could have impacted the completeness and accuracy of the data. This finding highlights the need for effective communication and trust-building with research teams, as well as the importance of effective data security measures in these and other platforms.

This study demonstrates the feasibility and acceptability of an audio-recording pipeline for speech-based cognitive assessment among older Black adults in HHC settings. Although recruitment was hindered by high medical complexity among nonparticipants, those who enrolled reported minimal disruption from audio recording and expressed positive attitudes toward the research. These results highlight the promise of speech-based tools for early MCI-ED detection and underscore the importance of inclusive recruitment strategies and technology safeguards, particularly for medically vulnerable populations.